Improve performance on ManyToMany chained filters

[bruno-fs]

I was experiencing the issue @eduardocalil reported on #745. At the time, the fix proposed by him wasnt accepted because:

1. it broke a test.
2. because the problem is more a django ORM issue than django-filter.

I can't comment on 2, but at least 1 dont block this PR anymore. The error was caused by currently unsupported versions of DRF.

[bruno-fs]

After inspecting the tests I realized they are not actually running.

[rpkilby]

After inspecting the tests I realized they are not actually running.

Can you point out which tests aren't running?

[carltongibson]

Hi.

This isn't a performance issue. The issue is that over M2M relations, this:

Blog.objects.filter(entry__headline__contains='Lennon', entry__pub_date__year=2008)

is not semantically equivalent to this:

Blog.objects.filter(entry__headline__contains='Lennon').filter(entry__pub_date__year=2008)

as documented in the Queries topic doc.

The solution here is NOT to get involved in ORM Dark Magic™.

Rather, you need to be creating a (multi-)filter that takes the (e.g. two) parameters and applies them in a single step.

The addition to Django Filter for this is:

• Document the pattern
• Add a filter class that allows declarative declaration.

I'll leave this open for now as a reminder.

[bruno-fs]

@rpkilby about the tests being skipped, it seems that all of them were skipped 😱
this happened not only on this PR but also on master.

[bruno-fs]

@rpkilby this is the change to make detox run and detect that I was breaking lots of tests

[rpkilby]

Yikes. This looks like a huge bug in tox/detox. Those exceptions should not result in passing builds.

@carltongibson - might be a good reason to enable the coverage status check?

[rpkilby]

I would recommend that we ditch detox in favor of regular tox. Honestly, I don't think we get much from parallelizing tox. The tests are pretty fast - most of the build time is just setting up the Travis container.

[carltongibson]

@rpkilby Are you good to make the PR for that?

[rpkilby]

Sure.

[rpkilby]

lol I never got around to making that PR. Done.

[rpkilby]

@carltongibson - I think you can close this now given the PR.

[carltongibson]

I left this open to document the pattern and/or add a multi-parameter filter base class or such.
(Have had ===0 time to look at that. 🙄)

[rpkilby]

Ah - thought you meant about the tox/detox thing. 👍

[carltongibson]

Moved documentation issue into #1020